recomendations for DSP development platform

[davygrvy]

Hi,

I'm looking to prove an idea, but I'm not sure a typical PC running a head-less linux distro would suffice.

I need to take multiple channels of input. For now, 8 discrete channels on AES/EBU, but that may change to HDMI later if possible without Dolby licensing rights. Output would be to AVB streams (possible 21 streams) using the services of OpenAVB along with a proper NIC such as an Intel I210.

I need to process the eight discrete channels for a multi-way 7.1 theater system doing a relatively complex matrix.

Off-board DSP engine?

[davygrvy]

Too bad the UAD-2 PCIe DSP Accelerator Cards are worthless under Linux

[bohdan1232000]

Originally Posted by davygrvy

Hi,

I'm looking to prove an idea, but I'm not sure a typical PC running a head-less linux distro would suffice.

I need to take multiple channels of input. For now, 8 discrete channels on AES/EBU, but that may change to HDMI later if possible without Dolby licensing rights. Output would be to AVB streams (possible 21 streams) using the services of OpenAVB along with a proper NIC such as an Intel I210.

I need to process the eight discrete channels for a multi-way 7.1 theater system doing a relatively complex matrix.

Off-board DSP engine?

Hi Dave,

I run 8in AES/EBU to 16Out AES/EBU DSP processor called Ultimate Equalizer. This is followed by miniDSP ICE-PWR AES/EBU amplifiers in various configurations, from 60W/8ohm to 250W/8ohm.

If this is sufficient for you, then you can find more information on http://www.bodziosoftware.com.au/

Best Regards,
Bohdan

[davygrvy]

Interesting.. How do I build one myself with open source software to run on Linux and replace the direct output to AVB streams? 16 output isn't enough. I'll need at least 21

[davygrvy]

For software, maybe BruteFIR along with DRC (?). But at 21 streams of deep filtering at 24b/96k I don't think I'll have enough horsepower unless I off-load the work. I'll almost sure of it

[davygrvy]

Oh, BruteFIR uses FFTW for the FFT engine. The CUDA toolkit from nVidia has a drop-in replacement called CUFFT for off-loading to the GPU.

rad!

[DPH]

I think you'll still be good on a modern processor. He (Anders) quotes this in the brutefir description:

With a massive convolution configuration file setting up BruteFIR to run 26 filters, each 131072 taps long, each connected to its own input and output (that is 26 inputs and outputs), meaning a total of 3407872 filter taps, a 1 GHz AMD Athlon with 266 MHz DDR RAM gets about 90% processor load, and can successfully run it in real time. The sample rate was 44.1 kHz, BruteFIR was compiled with 32 bit floating point precision, and the I/O delay was set to 375 ms. The sound card used was an RME Audio Hammerfall.

That processor is probably outpaced by most cell phone SOC nowadays, so I think you'll be good.

[davygrvy]

Like most things, I won't know until I try. But seeing there is an upgrade path, this looks like a solid choice and passes the sniff test.

CUFFT vs FFTW

CUFFT spanks FFTW only when the work to do is significant. Makes sense just like with multi-threaded programming. When the work to do isn't significant, the inter-thread communication, syncing and other properties of worker management takes precedence rather than the job.

One of these ought to do for taking the input.

I think I'm good to go.